Render nans as 'bad' in mpl cmaps when mode='colormap' #2427
Conversation
| @@ -137,7 +138,8 @@ def __call__(self, bounds=None): | |||
| data = interval(array) | |||
| data = contrast_bias(data, out=data) | |||
| data = stretch(data, out=data) | |||
| data[np.isnan(data)] = 0 | |||
| if self._set_nans_to_zero: | |||
| data[np.isnan(data)] = 0 | |||
There was a problem hiding this comment.
This was only introduced in fa60bf1 for compatibility with Astropy; if we can reproduce what the issue with Astropy was, preserving NaN may perhaps be made the default again.
There was a problem hiding this comment.
@astrofrog Any memory/tips of what we should check? Thanks!
There was a problem hiding this comment.
I think it was actually that astropy did not preserve NaNs and then started to so it's not an issue with astropy as such - but more that when you do color compositing you don't really have the option to give NaN values their own color since each channel has only one color assigned to it. So I think it's as I mentioned on the call - for multi-color compositing we should always set NaNs to 0 and for the colormap we could never set it to zero.
bmorris3
force-pushed
the
cmap-bad-values
branch
from
d85b0ee to
e01db76
Compare
|
As of e01db76, nans are forced to zero when |
bmorris3
changed the title
|
It looks like some tests will need to be updated |
bmorris3
force-pushed
the
cmap-bad-values
branch
from
e01db76 to
670aa34
Compare
|
@astrofrog All fixed! |
Description
By default, image viewers cast NaNs to zero before applying a colormap. This prevents the use of the matplotlib "bad" feature. For each colormap, values above or below the colormap vmin/vmax are rendered with an optionally specified value, and "bad" values (usually nans) are rendered by a third value:
By default,
badiscolor='k', alpha=0, but this can be changed with:which allows users to clearly visualize nans efficiently and without special handling.
Glue currently forces nans to zero before they are converted to colors by the colormap:
glue/glue/viewers/image/composite_array.py
Line 140 in 0f83a03
which prevents the user from ever making use of the
badfunctionality.This PR allows the user to set a private attribute on the
CompositeArrayclass called_set_nans_to_zero. When True, the behavior is unchanged. When set to False by a user, nans are not forced to zero, and thebadvalues are rendered as described by the mpl colormap. An example is shown below using Imviz from jdaviz: